ACAMS TODAY, December 2023–February 2024 December 2023–February 2024 | Seite 78

COMPLIANCE
ChatGPT and AI overview
One such example of an accessible AI application is ChatGPT , a program created by OpenAI . ChatGPT is one in a series of natural language models 4 deployed by OpenAI . For a layperson , language models might be comparable to a supercharged autocorrect engine . 5 That is to say , ChatGPT does not really know anything , but it is able to relay a series of words that are generally associated with each other .
Most significantly for all of us in the compliance sector , ChatGPT does not check facts . To the contrary , at the bottom of the ChatGPT prompt screen , there is this disclaimer : “ Free Research Preview . ChatGPT may produce inaccurate information about people , places , or facts .” Information provided by ChatGPT must be independently verified to be trusted . Overreliance on the accuracy of ChatGPT has already caused problems for legal professionals . One New York attorney was fined for submitting a legal brief that cited fictitious cases 6 compiled by ChatGPT , which has been known to invent cases or facts to cases . 7
The authors asked ChatGPT about itself and found that someone or something , most likely a human , wrote what amounts to a disclaimer so broad as to reduce or eliminate the use for AI . In part , it reads , “ ChatGPT sometimes writes plausible-sounding but incorrect or nonsensical answers .” 8 This disclaimer should be posted whenever text provided by ChatGPT is used , quoted or referenced in any form .
In short , ChatGPT uses machine learning algorithms to process and analyze data and generate responses to users ’ inquiries that are accessible , friendly and often couched in the first person , i . e ., “ Yes , I can ” 9 or “ However , I should note .” 10
This means that when users type a question into ChatGPT , it produces a conversational answer .
For example , when prompted to “ Describe how to use ChatGPT to prevent money laundering schemes in the United States ,” ChatGPT provided a list of ways it could be used to support a comprehensive AML program or regulatory compliance : “ ChatGPT can aid in preventing money laundering schemes in the United States by providing real-time analysis of financial transactions , identifying suspicious patterns or activities , and offering guidance on relevant regulations and compliance measures . By integrating GPT-3.5 into financial monitoring systems , it can help identify red flags , potential money laundering activities , and suggest appropriate reporting steps to relevant authorities . Additionally , it can provide training and educational resources to financial professionals to enhance their understanding of evolving money laundering tactics and regulatory updates , thus contributing to a proactive approach in combating financial crimes .” 11
Although it can be a helpful tool , ChatGPT will tell you that it is not a substitute for a comprehensive AML program , nor can it replace the human element .
ChatGPT ’ s limitations
ChatGPT has flaws : For one , ChatGPT is only as good as the information it has been fed . It is the old story of garbage in , garbage out since ChatGPT is limited by the veracity of its source data . In addition , it reflects the views back to us that it picks up from its source data , but not the internet at large . Unlike search engines like Google that have access to the entire internet , ChatGPT in its present form is limited to the source data available at the time it was created ( e . g ., data available up until September 2021 ). Further , AI is a reflection of societal knowledge and , therefore , is susceptible to the same things to which people are susceptible , such as sexism , 12 racial bias and political polarization . When you Google something , you are given multiple sources and it is up to you to trace what you want to use , from where the information is coming and who wrote it . When you ask ChatGPT something , you do not necessarily know who wrote it , let alone whether the information provided is accurate . This leads back to a flaw discussed above : ChatGPT has been known to make stuff up .
Effect on the AML field and on criminals
ChatGPT can provide accurate answers to a variety of AML and Bank Secrecy Act ( BSA ) -related issues . Below are questions asked to ChatGPT and the range of responses vary in accuracy .
Example : In response to being asked , “ What is the definition of a currency transaction report ?” the following appeared : “ A Currency Transaction Report ( CTR ) is a financial document required by the Financial Crimes Enforcement Network ( FinCEN ) in the United States , typically completed by financial institutions , which details any transactions involving cash amounts over a certain threshold ( currently $ 10,000 ) within a single business day . The report includes information about the transaction , the parties involved , and the purpose of the transaction , aiding in the detection and prevention of money laundering and other illicit activities .” 13
78 acamstoday . org